Hey y'all! Some of you may be familiar with me, others may not. For those that aren't: I'm an alumnus of Wisconsin and currently live in Madison. Because of that, I've been able to enjoy the fruits of Wisconsin's recent success: Rose Bowls (3-straight and counting!), 83 point blow-outs, Russell Wilson's angelic face. Unfortunately, as I grew up in Minnesota and didn't start cheering for the Badgers until 2007, I've not been able to enjoy the finer aspects of Wisconsin's history, like actually winning those Rose Bowls.
While I cheer for all of Wisconsin's sports (especially their most successful sport
women's hockey cross country) football is by far my favorite. Now, I consider myself reasonably knowledgeable about various aspects of the game, but I'm secure enough in myself to say that I'm woefully inadequate in several areas and prone to biases and/or hating anything relating to Michigan State. Because of this, I tend to heed to what numbers and statistics can tell me. It's my belief that the quality (or not) of a team's play will ultimately be supported in the box score, so I trust that over my own red-shaded eyes.
There are many wonderful resources out there for the statistically-minded, but because I like to try things out for myself, I've come up with my own system for evaluating college football teams. Is it valid? Well, I schooled everyone in the OTE Bowl Prediction Competition last year, so I'm going to go ahead and give myself some credit. What follows is a (semi) brief primer on what my ratings mean and give a short example of what my Big Ten season previews are going to look like.
DISCLAIMER: I am not an expert. I am not paid for this. I derive enjoyment out of doing these ratings, and sharing them with others is more fun than leaving them sit on my computer un-seen. If you want to see what actual professionals have to say, feel free to peruse other ends of the Internet, I will not be offended :)
Also, (tr)uck Michigan State.
I did this thing last year and went fairly in-depth on how I calculate these things. That was boring, so in the interest of... interest, I'll just give a relatively short explanation on how to interpret these metrics.
Adj Off - analogous to scoring offense, adjusted for opponent strength. Like scoring offense, the higher, the better. An average team has an Adj Off = 1.00. To give a sense of scale: last year, Nebraska had the best Adj Off in the Big Ten (1.65, 9th overall), Illinois had the worst Adj Off in the Big Ten (0.57, 123rd overall). Illinois did not have a good season, as you may have guessed.
Adj Def - analogous to scoring defense, adjusted for opponent strength. Like scoring defense, the lower, the better. An average team has an Adj Def = 1.00. To give a sense of scale: last year, Michigan State had the best Adj Def in the Big Ten (0.52, 6th overall), Indiana had the worst Adj Def in the Big Ten (1.34, 111th overall).
Adj Eff - analogous to scoring ratio, adjusted for opponent strength. Like scoring ratio, the higher, the better. An average team has an Adj Eff = 0.50. To give a sense of scale: last year, Michigan had the best Adj Eff in the Big Ten (0.64, 18th overall), Illinois had the worst Adj Eff in the Big Ten (0.38, 108th overall). Yes, the Big Ten was pretty bad last year.
Adj Marg - analogous to scoring margin, adjusted for opponent strength. Like scoring margin, the higher, the better. An average team has an Adj Marg = 0.00. To give a sense of scale: last year, Nebraska* had the best Adj Marg in the Big Ten (0.60, 14th overall), Illinois had the worst Adj Marg in the Big Ten (-0.38, 99th overall).
* Yes, shut up, I know. Ohio State was 16th if that helps alleviate anything.
Adj Comb Rating - Okay, a bit more of an explanation this time.
Adj Eff and Adj Marg are both decent measures of over-all team quality, however they both have their flaws. Adj Eff rewards teams with good defense, Adj Marg rewards teams with good offense. A team like Michigan State is looked upon much more favorably with Adj Eff than Adj Marg, a team like Nebraska is looked upon much more favorably with Adj Marg than Adj Eff. Neither rating is perfect on its own. In the past, I used Adj Marg as my "official" rating, because:
1) I believe that a good offense can overcome a bad defense more than the other way around (there is a lower limit to how many points a defense can give up, there is no such - theoretical - upper limit to how many points an offense can score)
2) It seemed to do better at predicting the outcome of games than Adj Eff. So I went with Adj Marg.
Still, I was dis-satisfied. I felt I wasn't giving enough credit to teams with good defense and I searched for a solution. So over the past winter I came up with Adj Comb which offered a compromise between Adj Eff and Adj Marg. However, since Adj Comb sounds stupid, I'm just going to refer to it as a team's overall Rating here. It's a simple linear combination of Adj Eff and Adj Marg*, and it's evaluated on the same scale as Adj Marg (average = 0.00, the higher, the better). Teams with good offenses are still looked up a bit more favorably (again, there's simply a limit on what a defense can do on the scoreboard), but it's much less egregious now. Nebraska still comes in at best in the Big Ten (0.55, 17th overall) and Illinois still comes in at worst in the Big Ten (-0.44, 103rd overall). For most teams, it's a minor change from their Adj Marg rating.
* 0.5 x (4 x AdjEff + AdjMarg - 2), if you're interested
OBLIGATORY FOLLOW-UP QUESTION: THE BEST ACCORDING TO THESE RATINGS. QUESTION MARK. - Well, according to my Excel sheets (with era-adjustments) it's 1966 Notre Dame (hush now Sparty fans). But honestly, I think it's war-time Army, without question. If I had to pick one, I'd say 1945 Army. They actually come in 3rd in my numbers (behind 1944 Army), but they had a ridiculous SOS - which they destroyed (Notre Dame went 7-2-1, came in 8th in the final AP poll... and lost 48-0 to Army. #3 Navy lost 32-13).
Resume - analogous to W/L record, adjusted for opponent strength. I hardly mention this at all, since I'm more interested in the Adj Off/Def stuff. But it pops up in some of my graphs, so I should explain what it is. It's my basic "BCS-style" rankings to offer a point of comparison to the Adj Off/Def rankings. Because most people and pollsters rank teams according to their W/L record with little regard to how close the games were, these are probably closer to what people think is a proper ordering of teams. Though, it can have hiccups as Notre Dame was the top-rated team for 2012 (hey, I'm not the only system that went that way).
Luck - Now it's time for the fun stuff! Now, "Luck" is kind of a nebulous concept, so I'm going to try and be as clear as I can be about this. "Luck" does not refer to injuries, bad officiating, turnovers, or fluky plays, though all of those are possible factors that can help explain a team's "Luck" rating. For my purposes, "Luck" is simply the difference between a team's actual winning percentage and their expected winning percentage (ActWinPct - ExpWinPct, in equation form). For most teams, this actually isn't a huge difference, the "Luck" works out to a difference of less than one game. Bill Parcells' sage knowledge still applies. For other teams (cough - Ohio State - cough), this difference can be significant, and works out to 2-3 games worth of "Luck." However, those are extreme cases.
So what do I mean by "expected winning percentage?" Well, say Team A is favored by 7 pts over Team B in Vegas. You'd expect Team A to win any individual game vs Team B, but if they played 10 times Team B might steal a couple, just because Team A isn't that much better than Team B. As I like to say: "Upsets happen y'all. That's why they call them upsets." So instead of saying "Team A is favored by 7 pts," you might say "Team A is 70% likely to win this game." The second gives a better description of the scenario.
To find a team's "expected record" then, just add up the confidence (percentage likelihood of winning) of each game on their schedule. Sure, coaching and match-ups play a role in winning games, but this is an effort to simplify and I'm a man with limited time. The question now becomes: from where do I get these confidence values? Go back to the Team A vs Team B hypothetical above. Team A was favored by 7 pts over Team B, I said that's worth a 70% chance of winning. That wasn't just me pulling numbers out of a hat, that's what it actually is. Of course, I'm not going to look into Vegas odds every week, so the conversion I use is based on my own ratings. Fortunately, I'm not too far off of Vegas. As illustrated by this plot:
The black data points are based on data going back to the 1940 season. The gray line is the quadratic regression that best matches that data. The red line is what I actually use in my calculations and was based on an earlier regression when I didn't have as much historical data. I'm not really worried about the difference between the two because:
1) The difference is really small and only matters at really high confidence values.
2) The further in the past you go, the more stuff gets weird. I'm more concerned with using these for recent football happenings.
3) It would be a lot of work to change everything. NOTE: the only real reason.
OF INTEREST: The small discrepancy at ExpDiff = 1.5 pts. I like to refer to this as the "Bill Snyder-Bret Bielema Coaching Valley." I feel the name is self-explanatory.
Obvious follow-up: where does the "expected difference" come from? From this equation:
ExpDiff = AvgPts x (Rating_A - Rating_B)
NOTE: I'd format these equations better, but SB Nation doesn't seem to have the HTML formatting set-up. Let me know if I'm missing something.
ExpDiff, Rating_A, Rating_B: I hope those are clear in what they mean. AvgPts is the FBS-wide pts/game scoring average. A good rule of thumb is 25 pts here, but it's calculated on a season-by-season basis and scoring is going up. Hey, another plot!
The y-axis is pts/game scoring average for the specified season, just to be clear.
Does that plot really explain anything? Not really, but I think it's kind of cool to look at. Honestly, some of the greatest joy I get in these ratings is seeing trends throughout history and how the college football landscape has changed over time.
Anyway, back to the equation. I'd like to say that I arrived at it through rigorous testing but... I made it up. I've been fortunate that it actually works decently, and I haven't been able to find anything consistently better using the data I have, but it's purely of my invention. To offer some support, I have done some testing. Assuming that my ExpDiff is an accurate predictor of what the actual scoring margin (ActDiff, we'll call it) is, I should be able to do a regression of ExpDiff vs ActDiff for all games in a season and arrive at:
ActDiff = m_scr x ExpDiff
With m_scr = 1.00 (or pretty close, at least). Fortunately, I have modest skills with MATLAB and have been able to do this for all the seasons I have on file (1940-present). The results are presented here:
sorry about the size, I did my best...
While there is a clear trend there, I'm still pretty happy because:
1) As I said before, the more you go in the past, the weirder stuff gets. Since the mid-1990's, things are pretty well centered around 1.00.
2) Even m_scr = 1.10 is a difference of only 2 pts at ExpDiff = 20 pts. Not enough to worry about.
If that trend keep up in the next few years, I'll revisit this. But I'm sticking to what I've got right now. Again, I schooled y'all in the bowl predictions.
Okay! I think that's all I have for getting y'all acquainted with what I like to talk about so you won't be confused. As I stated above, I like to make all these ratings, but I also like to share them with other people. What's the fun in doing stuff just by yourself (insert masturbation joke here)? Since it's the pre-season, the appropriate action is to make season projections and predictions. As luck would have it, numbers lend themselves well to doing those sorts of things. It is also helpful that this is college football and saying stuff like:
"Alabama will probably be pretty good. ROLL TIDE."
"Indiana won't make the Rose Bowl."
"Florida State will lose a game to some ACC mediocrity."
gets you half-way there already. So I'm being graded on a generous curve. Still, this is the sporting dead-season (No one cares about baseball, let's be honest. Pipe down you Detroit Tigers fans), so we need to fill it with something. I'll be doing previews for all the Big Ten teams (probably not Maryland or Rutgers, because I still don't want them here), and to give you an idea of what to expect, here's a team that's basically Big Ten anyway: Notre Dame (also, they give a good range for examples).
DRINKING FIGHTING IRISH
Notre Dame Overall Ratings
Notre Dame Adj Off/Def
Notre Dame Win Pct/Luck
As you may have gathered, the names below the Overall Ratings graph are a chronology of the coaches Notre Dame has had since 1940. In cases of interim coaches, this may be somewhat off of what you might expect. It would come down to which coach had the better winning percentage during his portion of the season. This is not my choice, this is simply the order of listing that Sports-Reference uses and I had no desire to try and "correct" this.
I use the convention of "F. LASTNAME," because Excel is weirdly terrible at their formatting option for bar graphs, and some coaches have names that were just too damn long to fit neatly.
The stars (actually asterisks, but close enough) at the bottom of the "Overall Ratings" graph indicate national titles, according to the AP, Coaches/UPI, FWAA, and (prior to 1935) Helms polls. Why? My decision, and Ohio State can accept their non-prize for 1970. Poll results gathered from Sports-Reference. Also, Minnesota should be deeply ashamed for their 1960 title.
I have data through 1940 here. That may change with the Big Ten previews, it depends if I get through the 1930's or not. Not that it matters much, but just so you know.
I think it's pretty well indicated, but I use percentiles here instead of the actual ratings. It's much easier to get a sense of how good a team was in a given year using this, especially for y'all who haven't spent multiple hours on this project. "Notre Dame was in the 30th percentile in 2007" vs "Notre Dame had a -0.32 Adj Marg rating in 2007." Ya know, like that.
These are honestly my favorite set of graphs that I produce. It's one thing to know tidbits about college football history, it's another to see it all displayed for you. Nebraska, in particular, is pretty interesting to see.
Projected Rating: 0.624 (11th)
|Conf||Proj Rec||Conf Rec||Adj Off||rk||Adj Def||rk||Adj Eff||rk||Adj Marg||rk|
|Schedule||SOS||rk||Conf SOS||rk||n-con SOS||rk|
|Week 1||Week 2||Week 3||Week 4||Week 5||Week 6||Week 7||Week 8|
|89 - Temple||97%||27 - FVCK-M||69%||70 - ?????||93%||30 - Sparty||72%||4 - Okie||38%||28 - ArizSt||70%||0 - BYE||0%||10 - SoCal||50%|
|Week 9||Week 10||Week 11||Week 12||Week 13||Week 14||Week 15||Week 16|
|73 - PLANES||93%||63 - BOATS||90%||41 - Pitt||77%||0 - BYE||0%||25 - Mormons||68%||12 - Stanford||50%||0 - BYE||0%||0 - BYE||0%|
Wow, that looks terrible. I'll try to find something a bit more pleasing to the eye before I start my Big Ten previews.
The first table should be pretty clear. Those are the projected ratings and records for Notre Dame. Surprise! Notre Dame should have a good defense and a lesser quality offense. Personally, I feel they will be better on offense than this projection. Everett Golson can only get better from his freshman year. Probably.
The second table has the schedule strength information. It's somewhat incomplete, as the Mountain West has yet to release their schedule (which is super annoying, by the by), but with Notre Dame, it doesn't really matter. Schedule strength is simply the average rating of all opponents. This is somewhat flawed (Is Alabama and New Mexico State tougher than Purdue and Iowa State? Kind of depends), but easy to calculate. So there.
The third table has the schedule, with opponent rankings and win projections. Happily, my color-coding seems to have copied over from Excel. Un-happily, it still looks kind of crappy. Still, I think it's pretty informative.
Notre Dame profile
EDIT: And I'm just now realizing I had a mistake in my non-conference calculations. It should be fixed now.
This last part is an attempt to give a graphical representation of a team's place in the FBS heirarchy. Think of it like a "Championship Profile." You want to be maxed out in all areas, those would be indicators of a championship-quality team. Obviously, Notre Dame is not the best example here for conference-only comparisons, but that's what the right-most plot is for. Considering the poor quality of the Excel table importing here, these graphs should serve as a good quick glance for the upcoming season. Like in the historical plots above, offense, defense, and overall rating are evaluated on a percentile basis (as well as SOS).
Okay! I think that's good enough. Again, I do this for fun, and I welcome any and all criticism or suggestions (especially as relates to HTML formatting on SB Nation here, geez). Y'all have a great week and remember: Sparty is the worst.